Making mushroom girls! - Trying out MeinaMix model checkpoint for Stable Diffusion. Many beautiful results.
I downloaded a new SD checkpoint yesterday called MeinaMix. I think it has the best anime results I've gotten so far on Stable Diffusion. I have tried PastelMix, Kenshi, AbyssOrangeMix3 (this one reminds me of an improved PastelMix). They all produce great images but none are as good as MeinaMix for concept art type illustrations, especially regarding high-quality ambiance, lighting and facial qualities.
The generation of mushroom girls
Not the easiest task. It took me over half an hour to even get something of decent quality. If careless and without position prompting (full-body, close-up, etc.), as well as high-res fix (I can explain in a more technical post later), there is no face visible, everything is too small, distorted and simply bad/unusable.
The results
After realizing what I needed to do, it took me 40 more minutes to get to good results by engineering toward my goal. Most images either didn't have the proper shroomie hat, or they were too undressed (some will say 'no such thing' lol), had wrong body features, had inconsistent floating objects, etc., but this one is good. AI art may be faster than traditional art, but it still takes a lot of effort to get what you want if it's complex and you're demanding.
Full body
I made dozens upon dozens of full-body pictures, maybe even over a hundred, because these were the base prompt, and my initial goal.
The following image is my absolute favorite from all the far-away full-body shroomie prompts. I'm sure I could perfect things, but this is good enough for a "test" run (lasting over 2 hours).
Close-up
Having praised this model so much, I couldn't leave after posting a poor-quality face. If the face is distant, I think all models will have trouble with it. I decided to make some close-ups to give the model a chance to output some high-quality faces to show off.
Normal face, medieval hat, shrooms in hand
Even closer, lacy hat made of shrooms
I'm 99% sure the painting style & bright yellow features are courtesy of WLOP. It's very similar in style to his. (link to a piece that showcases his typical earrings)
I LOVE these clothes and colors, sadly no shrooms
And what would a shroomgirl post be without a proper shroom hat.
Conclusion & Inspiration
Overall, I'm very satisfied with MeinaMix and I'm sure I'll be using it for some of my image creations on Stable Diffusion. I'll make a couple of more technical posts later explaining how/when/what/why I use Stable Diffusion.
I was very inspired by the following post:
[Exploring prompts and styles] Zoe and Me by @lucianaabrao
recently I testing out different prompts and exploring various styles and decided to play around and see what kind of creations I could come up with.
Decided to try something more personal and create an image of myself and my little bear Zoe.
Leave some love for her!
Ask any questions you have about SD, AI art or any related topic and I'll make sure to respond promptly,
Here’s my ❤️. Vibrant, colourful and we blended.
What do you mean by "we blended"?
Thank you for the ❤️ :) Some ❤️ for you too. Glad you like them :) I agree that the colors are beautiful!
bonus: some miniatures I liked
(the ones I posted are 1024x1024, these ones below are from the pre-process that generates 512x512 images)
bonus 2: some of the first discarded attempts
Nice!!!!
What a great style in terms of the model. The outtakes are brilliant, too. I've got so many models downloaded on my PC that it now gives me a bit of decision paralysis as to what I'm actually going to work with for any given idea that is flying around in my head on any given day.
Understand you... Some times few options to choose is better than a lot.
Yes, I always, always end up over producing images when using various prompts and compositional ideas.
Same hahah, I have so many thousands of discarded images. I don't delete them because I have like a dozen terabytes available but it's getting hard to even browse them or find the ones I want.
Adobe Bridge used to be great for culling images.
Hadn't thought of it. I thought it's paid software. It used to pop up all the time on my pirated copy of lightroom and I always ignored it for some reason, can't remember why, maybe I just didn't find much use for it? Idk. It was like 7+ years ago.
Yeah, I can't remember, and don't use it anymore. The cost of storage is so much lower now it doesn't really matter. It is just pain when you do the initial cull down.
Personally I kinda disagree. I try all viable models all the time for prompts I like, and choose the model with the best outputs to test prompt derivations. It's not one or the other but always the best and most apt for the task. Some are good for backgrounds, some for anime characters, some for photorealistic characters, some for sci-fi, some for fantasy, etc.
Most of the time, though!
Sometimes I'd just like to go to midjourney because I want a fast image and don't care if it's got that perfect anime style or that perfect dark 60s photography feel. Which reminds me I wanna pay for midjourney but it's so expensive. Sadsad.
I need to pay for it too.
Accepting hive already? 😆
Good, understand, but i have some difficult choosing among too many options... But for different works is needed
I would but I don't do anything that justifies the expense :( I'm not a professional artist, and the cost of MJ is very high compared to my art income. I have never made more than what it costs in returns for art. Maybe if I get a good job later this year :)
Kinda! I have been on hive since 2017 and I have never browsed it much daily for content. I'm always going for places like Reddit where things are neatly filtered and rated by category & quality, and I follow some categories.
When the communities hit in, I had already gotten burned out of Hive, and I'm only now trying to come back, so it will take some time & effort to take in all the new things and learn to enjoy this way of browsing.
I hope I accept it and flourish into it soon :)
Maybe it's because I'm very patient (some people suffer looking at me willingly fail over and over hahah, or failing with me). I'm okay with losing a few hours in the process without anything good, just to know what doesn't work. I enjoy the process of seeing and filtering out all the failure to take out the few successes and roll with them to victory.
With SD, I pick around 7 models, I put my prompts there, I hit play on all of it, and it can take 30-50 minutes to finish, so I go and brush my teeth, wash the dishes, clean my room, then I come back and see which ones are viable. If nothing is, I try again with more prompts. Sometimes I watch Netflix while I wait, and 99% of the results get discarded. Some days I don't share any of the results I spent hours creating.
It sounds absolutely bonkers even to myself which is why I have stopped doing it as much as I would before (I spent months working more than 10 hours a day on this lol). Nowadays I feel like doing some cool creative posting so I'm doing a temporary exception, but in the long term I might either learn some time-saving tricks or go with Midjourney because spending 4 hours a day to get $5-7 if lucky per post is not a viable business idea :)
Understand you.
I used to purchase stock images for my jobs. Now, i still purchase, but sometimes i want a specific image that is hard to find. So, MJ help getting the idea out of my head. Otherwise i have to select similar ideas from stock images.
Thanks for the mention.
You are doing great! Remember to save the prompts you liked and keep making variations until you get the wanted result.
This way you will have some best prompts ready that match to the art style you want.
This is what i do ♥_♥! I have a big folder with tens of thousands of images and each of them have a prompt. I want to build a system to browse them hahah cuz it's getting hard. These shroomies for example come from a previous prompt I used to make mushroom houses.
Beautiful style. Reminds me a little alien world.
Wow, this was really awesome in results! @ausbitbank might be interested in this model! Excellent work!
:D I love that despite how much effort and time it took, it is something that you like. I hope to make more good images :)
I appreciate your appreciation.
Next time, if you want, add those smaller ones into your post as well. These did all turn out pretty amazing! Thanks for sharing! Looking forward to seeing more of your outputs. I'm sure you probably already know... but if you find images you really like and have access to the seed... You can keep up with and reuse the seed so you don't have too much of a variance on the next run.
I still don't know much about the influence of seeds. I do reuse seeds for the same prompt or variations of it, but for completely new prompts I've never tried reusing an old seed because I don't know if there will be any advantage to it. I tend to just leave things to work 20-60 times and grab the best results and make bigger and more detailed derivatives, I go for 15-30 of those and then post the best ones.
Do you know more about what the seeds do if the prompt is completely different?
The seed will keep a similar style to what you have previously seen as the results. So, if you do a completely different prompt, then you might get some similar primary features or a similar style, even though the prompt is different. You will have to play around with the prompts with a same seed. Like do three different prompts with a seed that has produced results you like prior to running your test runs.
Seeds, to me, are the biggest mystery of this whole process. I have no idea where the seeds come from or why using the same one creates some continuity in a baseline... but once you know, the same seed creates similar results... then whe you find good results, perhaps save that seed and try it out again on completely different prompts and see what you can come up with!